Business scene determination method and device, computer equipment and storage medium

Abstract

The invention relates to a business scene determination method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining voice information of a user, and converting the voice information into text information; respectively matching the text information with scene information corresponding to each candidate service scene to obtain a first scene matching value corresponding to each candidate service scene; wherein the first scene matching value comprises a keyword matching value and / or a semantic matching value; based on a scene classification model, determining a second scene matching value corresponding to each candidate service scene according to the text information and the service information of the user; and selecting a target service scene corresponding to the user from the candidate service scenes according to the first scene matching value and the second scene matching value corresponding to each candidate service scene. According to the scheme, in the process of identifying the business scene, the keywords, the semantics and the business information are fused, and compared with business scene identification only according to the keywords, the business scene identification accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of business scenario recognition, and in particular to a business scenario determination method and device, computer equipment and a storage medium. BACKGROUND

[0002] With the improvement of service refinement in the telecommunications industry, semantic analysis of the conversation between service personnel and users is crucial for improving service quality and mining business opportunities.

[0003] In the traditional technology, the voice signal of the conversation is converted into text through automatic speech recognition (ASR), and then keyword matching is performed based on preset keywords (such as "disassembly" and "upgrade package") to determine the business scenario.

[0004] However, pure keyword matching is prone to misjudgment of related expressions as core intentions, and there is a problem of low accuracy of business scenario recognition. SUMMARY

[0005] Therefore, it is necessary to provide a business scenario determination method, device, computer equipment and storage medium to improve the accuracy of business scenario recognition.

[0006] In a first aspect, the present application provides a business scenario determination method, comprising:

[0007] obtaining voice information of a user and converting the voice information into text information;

[0008] matching the text information with scenario information corresponding to each candidate business scenario respectively to obtain a first scenario matching value corresponding to each candidate business scenario; wherein the first scenario matching value comprises a keyword matching value and / or a semantic matching value;

[0009] determining a second scenario matching value corresponding to each candidate business scenario based on a scenario classification model and the text information and business information of the user;

[0010] selecting a target business scenario corresponding to the user from the candidate business scenarios according to the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario.

[0011] In one embodiment, matching the text information with scenario information corresponding to each candidate business scenario respectively to obtain a first scenario matching value corresponding to each candidate business scenario comprises:

[0012] extracting keywords in the text information;

[0013] For each candidate business scenario, the extracted keywords are matched with reference keywords corresponding to the candidate business scenario to obtain a matching result.

[0014] According to the matching result, a keyword matching value corresponding to the candidate business scenario is obtained.

[0015] In one of the embodiments, the text information is matched with scenario information corresponding to each candidate business scenario respectively to obtain a first scenario matching value corresponding to each candidate business scenario, including:

[0016] A first text vector corresponding to the text information is obtained.

[0017] For each candidate business scenario, a similarity between the first text vector and a scenario feature vector corresponding to the candidate business scenario is determined.

[0018] According to the similarity and the keyword matching value corresponding to the candidate business scenario, a semantic matching value corresponding to the candidate business scenario is determined.

[0019] In one of the embodiments, based on a scenario classification model, a second scenario matching value corresponding to each candidate business scenario is determined according to the text information and business information of the user, including:

[0020] The business information of the user is obtained.

[0021] A second text vector corresponding to the text information and a business vector corresponding to the business information are fused to obtain a fusion vector.

[0022] The fusion vector is input into the scenario classification model to obtain a second scenario matching value corresponding to each candidate business scenario output by the scenario classification model.

[0023] In one of the embodiments, the target business scenario corresponding to the user is selected from the candidate business scenarios according to the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario, including:

[0024] For each candidate business scenario, a first scenario matching value and a second scenario matching value corresponding to the candidate business scenario are fused to obtain a fusion scenario value.

[0025] The candidate business scenario corresponding to the highest fusion scenario value is determined as the target business scenario corresponding to the user.

[0026] In one of the embodiments, the first scenario matching value and the second scenario matching value corresponding to the candidate business scenario are fused to obtain the fusion scenario value, including:

[0027] Obtain the first target weight corresponding to the keyword matching value, the second target weight corresponding to the semantic matching value, and the third target weight corresponding to the second scene matching value;

[0028] The keyword matching value is weighted using the first target weight, the semantic matching value is weighted using the second target weight, and the second scene matching value is weighted using the third target weight.

[0029] Based on the weighted processing results, the fusion scenario value corresponding to the candidate business scenario is determined.

[0030] In one embodiment, obtaining the first target weight corresponding to the keyword matching value, the second target weight corresponding to the semantic matching value, and the third target weight corresponding to the second scene matching value includes:

[0031] Obtain the first basic matching weight corresponding to the keyword matching value, the second basic matching weight corresponding to the semantic matching value, and the third basic matching weight corresponding to the second scenario matching value;

[0032] Based on the deviation between the keyword matching value and the keyword reference value, the first basic matching weight is corrected to obtain the first target weight;

[0033] Based on the deviation between the semantic matching value and the semantic matching reference value, the second basic matching weight is corrected to obtain the second target weight;

[0034] Based on the deviation between the model's predicted value and the model's predicted reference value, the third basic matching weight is corrected to obtain the third target weight.

[0035] Secondly, this application also provides a business scenario determination device, including:

[0036] The acquisition module is used to acquire the user's voice information and convert the voice information into text information;

[0037] The matching module is used to match the text information with the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value corresponding to each candidate business scenario; wherein, the first scenario matching value includes keyword matching value and / or semantic matching value;

[0038] The determination module is used to determine the second scenario matching value corresponding to each candidate business scenario based on the scenario classification model, according to the text information and the user's business information;

[0039] The selecting module is configured to select a target service scenario corresponding to the user from the candidate service scenarios according to the first scenario matching value and the second scenario matching value corresponding to each candidate service scenario.

[0040] In a third aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0041] obtaining voice information of a user and converting the voice information into text information;

[0042] matching the text information with scenario information corresponding to each candidate service scenario respectively to obtain a first scenario matching value corresponding to each candidate service scenario; wherein the first scenario matching value comprises a keyword matching value and / or a semantic matching value;

[0043] determining a second scenario matching value corresponding to each candidate service scenario based on a scenario classification model and according to the text information and service information of the user;

[0044] selecting a target service scenario corresponding to the user from the candidate service scenarios according to the first scenario matching value and the second scenario matching value corresponding to each candidate service scenario.

[0045] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program implements the following steps when executed by a processor:

[0046] obtaining voice information of a user and converting the voice information into text information;

[0047] matching the text information with scenario information corresponding to each candidate service scenario respectively to obtain a first scenario matching value corresponding to each candidate service scenario; wherein the first scenario matching value comprises a keyword matching value and / or a semantic matching value;

[0048] determining a second scenario matching value corresponding to each candidate service scenario based on a scenario classification model and according to the text information and service information of the user;

[0049] selecting a target service scenario corresponding to the user from the candidate service scenarios according to the first scenario matching value and the second scenario matching value corresponding to each candidate service scenario.

[0050] In a fifth aspect, the present application provides a computer program product, comprising a computer program, and the computer program implements the following steps when executed by a processor:

[0051] obtaining voice information of a user and converting the voice information into text information;

[0052] The text information is matched with the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value for each candidate business scenario; wherein, the first scenario matching value includes keyword matching value and / or semantic matching value;

[0053] Based on the scenario classification model, a second scenario matching value is determined for each candidate business scenario according to the text information and the user's business information.

[0054] Based on the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario, the target business scenario corresponding to the user is selected from each candidate business scenario.

[0055] The aforementioned business scenario determination method, apparatus, computer equipment, and storage medium acquire user voice information and convert it into text information. The text information is then matched with scenario information corresponding to each candidate business scenario to obtain a first scenario matching value for each candidate business scenario. This first scenario matching value includes keyword matching values ​​and / or semantic matching values. Based on a scenario classification model, a second scenario matching value is determined for each candidate business scenario according to the text information and the user's business information. Finally, based on the first and second scenario matching values ​​for each candidate business scenario, the target business scenario corresponding to the user is selected from the candidate business scenarios. This scheme integrates keywords, semantics, and business information during the business scenario identification process, improving the accuracy of business scenario identification compared to identifying business scenarios solely based on keywords. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart illustrating a method for determining a business scenario in one embodiment;

[0058] Figure 2 This is a schematic diagram of the process for obtaining the first scene matching value in one embodiment;

[0059] Figure 3 This is a schematic diagram of the process for obtaining the first scene matching value in another embodiment;

[0060] Figure 4 This is a schematic diagram of the process for obtaining the second scene matching value in one embodiment;

[0061] Figure 5 This is a schematic diagram of the process for obtaining fused scene values ​​in one embodiment;

[0062] Figure 6 This is a schematic diagram of the process for obtaining the weights of each matching value for weighted calculation in one embodiment;

[0063] Figure 7 This is a flowchart illustrating the business scenario determination method in another embodiment;

[0064] Figure 8 This is a structural block diagram of a business scenario determination device in one embodiment;

[0065] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0067] The business scenario determination method provided in this application embodiment can be applied to application scenarios where telecommunications business scenarios are identified based on the voice dialogue between business hall staff and users.

[0068] This method can be executed by a server or a terminal. The server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle systems, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses.

[0069] In one exemplary embodiment, such as Figure 1 As shown, a method for determining business scenarios is provided. Taking the application of this method to a server as an example, the method includes the following steps:

[0070] S101, acquires the user's voice information and converts the voice information into text information.

[0071] Voice information refers to audio data generated during communication between users and telecommunications service personnel, such as recordings of conversations between users and sales staff in business halls, and audio communication during installation and maintenance services.

[0072] For example, audio recordings of conversations between users and service personnel can be collected using data collection terminals such as name tags and recording devices carried by installation and maintenance personnel deployed in the service hall. The collected audio data is then encrypted and transmitted to a server. Furthermore, automatic speech recognition technology can be used to transcribe the audio data, converting the speech signal into a processable text format to obtain text information. For instance, if a user says in the service hall, "My broadband has been frequently disconnecting lately, I'd like to know how to fix it," the corresponding text information can be obtained after automatic speech recognition and transcription.

[0073] S102, match the text information with the scenario information corresponding to each candidate business scenario to obtain the first scenario matching value corresponding to each candidate business scenario.

[0074] The candidate business scenarios are a pre-defined set of common business types in the telecommunications industry, including but not limited to broadband disconnection, package upgrade, fault reporting, and tariff consultation.

[0075] The scenario information consists of feature data associated with each candidate business scenario, including the set of keywords and semantic feature descriptions for that scenario.

[0076] For example, a scenario information database for each candidate business scenario can be pre-built, and a corresponding set of reference words and semantic feature descriptions can be compiled for each scenario. Then, the text information can be compared with the scenario information of each candidate business scenario, and the degree of association can be calculated from at least one dimension of keyword matching and semantic matching to obtain a first scenario matching value. The first scenario matching value is used to measure the degree of association between the text information and the candidate business scenario.

[0077] The first scenario matching value can be a keyword matching value, a semantic matching value, or both. The keyword matching value is calculated based on the matching of keywords extracted from the text information with reference words for the candidate business scenario. The semantic matching value is calculated based on the similarity between the overall semantics of the text information and the semantic features of the candidate business scenario.

[0078] S103, based on the scenario classification model, determines the second scenario matching value corresponding to each candidate business scenario according to the text information and the user's business information.

[0079] Among them, the scene classification model is a machine learning model that is trained on telecommunications corpus and can combine text information and user business information to make scene judgments. For example, the scene classification model can be an improved model of the Bidirectional Encoder Representations from Transformers (BERT) model.

[0080] For example, user business information can be collected first, such as user package type, historical complaint records, user value rating, etc., which can be retrieved from the telecommunications business system.

[0081] Then, the text information is converted into text vectors, and the user business information is converted into business vectors. The text vectors and business vectors are fused and input into a pre-trained scene classification model. The scene classification model outputs the matching probability of the text information with each candidate business scene, which is the second scene matching value.

[0082] S104. Based on the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario, select the target business scenario corresponding to the user from each candidate business scenario.

[0083] For example, the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario can be comprehensively calculated. For example, a weighted summation method can be used to select the candidate business scenario with the highest value based on the calculation results, and determine it as the target business scenario corresponding to the user.

[0084] The aforementioned method for determining business scenarios involves acquiring the user's voice information and converting it into text. The text information is then matched against the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value for each candidate scenario. This first scenario matching value includes keyword matching values ​​and / or semantic matching values. Based on a scenario classification model, a second scenario matching value is determined for each candidate business scenario according to the text information and the user's business information. Finally, based on the first and second scenario matching values ​​for each candidate business scenario, the target business scenario corresponding to the user is selected from the candidate business scenarios. This scheme integrates keywords, semantics, and business information during the business scenario identification process, improving the accuracy of business scenario identification compared to identifying business scenarios solely based on keywords.

[0085] In some alternative implementations, the first scenario matching value can be a keyword matching value; based on this, see [link to relevant documentation]. Figure 2 , Figure 2 A flowchart for obtaining the first scene matching value is provided, which includes the following steps:

[0086] S201, Extract keywords from the text information.

[0087] Among them, keywords are words extracted from the text information that can reflect the user's core demands or the main content of the conversation, such as "high cost", "network outage", "equipment disassembly", "upgrade", etc.

[0088] Exemplarily, a Chinese word segmentation algorithm can be used to segment the text information, remove stop words such as "de", "le", "wo" that have no practical meaning, and screen out words that can express the core meaning as keywords. Among them, the Chinese word segmentation algorithm can be Jieba word segmentation, Han Language Processing (HanLP) word segmentation, etc., which can be selected according to actual needs.

[0089] Exemplarily, in order to improve the quality of the text information, a dialect mapping library can be pre-constructed. The dialect mapping library stores the corresponding rules between different dialects and standard Mandarin. For example, map "唔要宽带", "网线不要了", "断网" to "拆机". Furthermore, the text corresponding to the dialect can be mapped to the corresponding text information.

[0090] S202, For each candidate business scenario, match the extracted keywords with the reference words corresponding to the candidate business scenario to obtain a matching result.

[0091] Among them, reference words are a set of words preset for each candidate business scenario that can represent the core features of the business scenario, including core trigger words, auxiliary related words, exclusion words, etc.

[0092] Exemplarily, a reference word library can be constructed for each candidate business scenario. For example, taking the candidate business scenario of "equipment disassembly retention" as an example, in its corresponding reference word library, the core trigger words include "equipment disassembly", "cancellation of broadband", "network suspension", "don't want it", etc., with a weight of 3 points; the auxiliary related words include "too expensive", "cheaper", "network transfer", "mobile", etc., with a weight of 2 points; the exclusion words include "repair", "fault", "lag", "disconnection", etc., with a weight of -1 point, which is used to reduce misjudgment.

[0093] Furthermore, for each candidate business scenario, compare the extracted keywords with the reference words in the reference word library of this business scenario one by one, and record information such as the number of reference words matched and the weight of the matched reference words to form a matching result. The matching result is used to represent the overlap situation between the extracted keywords and the reference words of the candidate business scenario, including the number of reference words matched and the type of the matched reference words.

[0094] For example, the context window size can be preset. This means that auxiliary related words and excluded words must appear within the context window size of the core trigger word to be considered a valid match relevant to the current business scenario. The context window size can be set according to actual needs; for example, it can be set to 5 words.

[0095] For example, suppose a user says, "My previous plan was too expensive, and now my broadband keeps disconnecting. I want to get it fixed." The phrase "too expensive" (auxiliary word) and "get it fixed" (core word, fault reporting scenario) are too far apart. Without context window constraints, it might be mistakenly associated with the "plan downgrade" scenario. Context window can filter out this invalid association.

[0096] Suppose a user says, "I want to dismantle my phone because the current plans are too expensive." "Too expensive" (auxiliary word) is within the scope of "dismantle" (core word), so it can be confirmed as a valid reason for the "dismantle scenario" and can be included in the matching value calculation.

[0097] S203. Based on the matching results, obtain the keyword matching value corresponding to the candidate business scenario.

[0098] The keyword matching value is a quantified numerical value based on the matching results, used to intuitively reflect the degree of relevance between text information and candidate business scenarios at the keyword level. For example, matching extracted keywords with this reference thesaurus yields the core trigger word "cancellation" and the auxiliary related words "broadband" and "account," resulting in one core trigger word and two auxiliary related words. According to the pre-designed scoring rules, the keyword matching value is calculated as 1×3+2×1=5 points. Therefore, the keyword matching value for the "broadband disconnection" scenario is 5 points.

[0099] In the above embodiments, through a clear keyword extraction, matching, and scoring process, the degree of correlation between text information and candidate business scenarios at the keyword level can be quickly quantified, providing reliable basic data for the first scenario matching value. This process, by distinguishing the weights of core trigger words, auxiliary related words, and excluded words, makes the keyword matching value more reflective of the user's core intent, thus improving the accuracy of keyword matching.

[0100] In some alternative implementations, the first scene matching value can be a semantic matching value; based on this, see [link to relevant documentation]. Figure 3 , Figure 3 An alternative flowchart for obtaining the first scene matching value is provided, which includes the following steps:

[0101] S301, Obtain the first text vector corresponding to the text information.

[0102] For example, textual information can be input into a pre-trained vector encoding model. The model, through its understanding of the text's semantics, outputs a corresponding high-dimensional numerical vector, i.e., the first text vector. This first text vector represents the semantic features of the textual information. The pre-trained vector encoding model can be a sentence-level Transformer-based bidirectional encoder representation model (Sentence-BERT).

[0103] S302, for each candidate business scenario, determine the similarity between the first text vector and the scenario feature vector corresponding to the candidate business scenario.

[0104] The scenario feature vector is a high-dimensional numerical vector pre-constructed for each candidate business scenario, capable of representing the semantic features of that scenario. It is obtained by vector encoding the standardized descriptive text of the scenario. For example, a standardized description of the business scenario can be generated based on unstructured text such as historical Business Support System (BSS) work order data and operation manuals, and then encoded into a high-dimensional vector by the Sentence-BERT model to obtain the scenario feature vector corresponding to the candidate business scenario.

[0105] For example, for each candidate business scenario, the corresponding scenario feature vector can be retrieved, and the cosine similarity algorithm can be used to calculate the similarity between the first text vector and the scenario feature vector. The value of cosine similarity ranges from -1 to 1. The closer the value is to 1, the higher the semantic association between the two.

[0106] S303. Based on the similarity and the keyword matching value corresponding to the candidate business scenario, determine the semantic matching value corresponding to the candidate business scenario.

[0107] For example, a similarity weight coefficient α can be pre-set, with α ranging from 0 to 1, and can be adjusted according to the actual scenario. The keyword matching value is then normalized to obtain a normalized keyword matching value. The semantic matching value can then be calculated using the formula: "Semantic matching value = α × Similarity + (1-α) × Normalized keyword matching value". Here, normalization transforms the keyword matching value to a range between 0 and 1, aligning it with the similarity value range for easier weighted calculation.

[0108] In the above embodiments, text information is converted into high-dimensional vectors to capture the deep semantic features of the text, avoiding semantic associations that simple keyword matching cannot cover. Simultaneously, the semantic matching value is calculated by combining the keyword matching value with the semantic matching value, ensuring that the semantic matching result considers both the overall semantics of the text and the associations at the keyword level, further improving the accuracy of semantic matching.

[0109] In some alternative implementations, see [link to relevant documentation]. Figure 4 , Figure 4 A flowchart for obtaining the second scene matching value is provided, which includes the following steps:

[0110] S401, Obtain user's business information.

[0111] For example, relevant user business data can be queried and extracted from the system database by calling the interface of the telecommunications business system, including user package type, package usage duration, number of historical complaints, user value level, and business transaction records in the past three months.

[0112] S402, merge the second text vector corresponding to the text information and the business vector corresponding to the business information to obtain the fused vector.

[0113] The second text vector can be a high-dimensional numerical vector obtained by converting text information through a vector encoding model. It is used to represent the semantic features of the text in the scene classification model and can use the same or different encoding models as the first text vector.

[0114] A business vector is a high-dimensional numerical vector obtained by converting a user's business information. It can represent the user's business characteristics, such as the quantitative representation of information like package type and number of historical complaints.

[0115] For example, a vector encoding model can be used to convert text information into a second text vector. For user business information, categorical data (such as package type) can be one-hot encoded, and numerical data (such as historical complaint counts and user value levels) can be normalized. The one-hot encoding result and the normalization result can then be combined to obtain a business vector. Furthermore, a concatenation fusion method can be used to concatenate the second text vector and the business vector to obtain a fused vector that simultaneously contains textual semantic information and user business information.

[0116] S403, input the fusion vector into the scene classification model to obtain the second scene matching value corresponding to each candidate business scene output by the scene classification model.

[0117] For example, the fused vector can be input into a pre-trained scene classification model. After analyzing and processing the fused vector, the model outputs the matching probability corresponding to each candidate business scene. This probability is the second scene matching value. The value of the second scene matching value ranges from 0 to 1, with a higher value indicating a higher degree of matching.

[0118] In the above embodiments, by integrating textual information and user business information, the scene classification model can understand user intent from multiple dimensions, avoiding the limitations of relying solely on textual information for scene judgment. User business information reflects the user's historical behavior and current business status, providing an important reference for scene judgment and improving the accuracy of the second scene matching value.

[0119] In some optional implementations, the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario can be fused, and the target business scenario corresponding to the user can be determined based on the fusion result.

[0120] For example, for each candidate business scenario, the first scenario matching value and the second scenario matching value corresponding to the candidate business scenario can be fused to obtain a fused scenario value. For example, the first scenario matching value and the second scenario matching value can be weighted to obtain a fused scenario value; or, the first scenario matching value and the second scenario matching value can be input into a fusion model for fusion processing to obtain a fused scenario value.

[0121] Therefore, the candidate business scenario corresponding to the highest fusion scenario value can be identified as the target business scenario for the user.

[0122] For example, see Figure 5 , Figure 5 A flowchart for obtaining fused scene values ​​is provided, which includes the following steps:

[0123] S501, obtain the first target weight corresponding to the keyword matching value, the second target weight corresponding to the semantic matching value, and the third target weight corresponding to the second scenario matching value.

[0124] The first objective weight is a weighting coefficient applied to the keyword matching values, reflecting the importance of the keyword matching values ​​in the calculation of the fused scenario value. The second objective weight is a weighting coefficient applied to the semantic matching values, reflecting the importance of the semantic matching values ​​in the calculation of the fused scenario value. The third objective weight is a weighting coefficient applied to the second scenario matching values, reflecting the importance of the second scenario matching values ​​in the calculation of the fused scenario value.

[0125] For example, based on the characteristics of the business scenario and the results of historical data statistics, the initial weights of the keyword matching value, semantic matching value and the second scenario matching value can be preset in advance. Then, the initial weights of the keyword matching value, semantic matching value and the second scenario matching value can be dynamically adjusted in combination with the actual matching situation to obtain the first target weight, the second target weight and the third target weight, and the sum of the first target weight, the second target weight and the third target weight is 1.

[0126] For example, by analyzing the identification results of a large number of historical cases, the weights were adjusted to optimize the accuracy of business scenario identification. Finally, the weight of the first target was determined to be 0.2, the weight of the second target was 0.2, and the weight of the third target was 0.6.

[0127] S502, the keyword matching value is weighted using the first objective weight, the semantic matching value is weighted using the second objective weight, and the second scenario matching value is weighted using the third objective weight.

[0128] For example, the keyword matching value can be weighted using a first objective weight to obtain a weighted result, that is, the keyword matching value is multiplied by the first objective weight to obtain the keyword weighted value; the semantic matching value can be weighted using a second objective weight to obtain a weighted result, that is, the semantic matching value is multiplied by the second objective weight to obtain the semantic weighted value; and the second scene matching value can be weighted using a third objective weight to obtain a weighted result, that is, the second scene matching value is multiplied by the third objective weight to obtain the model output weighted value.

[0129] S503. Based on the weighted processing results, determine the fusion scenario value corresponding to the candidate business scenario.

[0130] Furthermore, the keyword weighted value, semantic weighted value, and model output weighted value can be added together, and the sum is the fusion scenario value corresponding to the candidate business scenario.

[0131] In the above embodiments, by assigning corresponding target weights to different matching values, the influence of each matching value in the fusion calculation can be adjusted according to the reliability and importance of each matching value, making the calculation of the fusion scenario value more targeted.

[0132] In some optional implementation scenarios, see Figure 6 , Figure 6 A flowchart illustrating the process of obtaining the weights for weighted calculation of each matching value is provided, specifically including the following steps:

[0133] S601, obtain the first basic matching weight corresponding to the keyword matching value, the second basic matching weight corresponding to the semantic matching value, and the third basic matching weight corresponding to the second scenario matching value.

[0134] The first basic matching weight is the initial weight coefficient preset for the keyword matching value. The second basic matching weight is the initial weight coefficient preset for the semantic matching value. The third basic matching weight is the initial weight coefficient preset for the second scenario matching value. The sum of the three basic matching weights is 1.

[0135] For example, based on business experience and preliminary experimental results, a first basic matching weight, a second basic matching weight, and a third basic matching weight can be set. For instance, the first basic matching weight can be set to 0.25, the second basic matching weight to 0.25, and the third basic matching weight to 0.5.

[0136] S602, based on the deviation between the keyword matching value and the keyword reference value, the first basic matching weight is corrected to obtain the first target weight.

[0137] The keyword reference value is a pre-set standard value used to measure the keyword matching effect, such as the median value of the optimal range of keyword matching values ​​obtained from historical data statistics.

[0138] For example, the deviation between the keyword matching value and the keyword reference value can be calculated. The deviation value can be positive, negative or zero. A positive value indicates that the actual matching value is higher than the reference value, and a negative value indicates that the actual matching value is lower than the reference value.

[0139] For example, a correction coefficient can be set based on the magnitude and direction of the deviation value. If the absolute value of the deviation is small, it indicates that the actual matching value is close to the reference value, and the correction to the basic weight is small; if the absolute value of the deviation is large, it indicates that the actual matching value deviates significantly from the reference value, and the correction to the basic weight is large. For instance, when the absolute value of the deviation between the keyword matching value and the keyword reference value is small, such as when the deviation is less than a preset threshold, the correction coefficient is 0.95, and the first target weight = first basic matching weight × correction coefficient; when the absolute value of the deviation is large, such as when the deviation is greater than or equal to the preset threshold, the correction coefficient is 0.8, and the first target weight = first basic matching weight × correction coefficient. The preset threshold can be set based on empirical data or actual conditions, and no specific limitations are made here.

[0140] S603, based on the deviation between the semantic matching value and the semantic matching reference value, the second basic matching weight is corrected to obtain the second target weight.

[0141] The semantic matching reference value is a pre-set standard value used to measure the effectiveness of semantic matching. The deviation between the semantic matching value and the semantic matching reference value can be calculated. The deviation value can be positive, negative, or zero. A positive value indicates that the actual matching value is higher than the reference value, and a negative value indicates that the actual matching value is lower than the reference value.

[0142] For example, when the absolute value of the deviation between the semantic matching value and the semantic matching reference value is small, for example, the deviation value is less than a preset threshold, the correction coefficient is 0.9, and the first target weight = first basic matching weight × correction coefficient; when the absolute value of the deviation is large, for example, the deviation value is greater than or equal to the preset threshold, the correction coefficient is 0.85, and the second target weight = second basic matching weight × correction coefficient. The preset threshold can be set based on empirical data or actual conditions, and is not specifically limited here.

[0143] S604. Based on the deviation between the model prediction value and the model prediction reference value, the third basic matching weight is corrected to obtain the third target weight.

[0144] The model prediction reference value is a pre-set standard value used to measure the model's prediction performance. The deviation between the model prediction value and the model prediction reference value can be calculated. The deviation value can be positive, negative, or zero. A positive value indicates that the actual matching value is higher than the reference value, and a negative value indicates that the actual matching value is lower than the reference value.

[0145] For example, when the absolute value of the deviation between the model's predicted value and the model's predicted reference value is small, such as when the deviation is less than a preset threshold, the correction coefficient is 0.9, and the first target weight = first basic matching weight × correction coefficient; when the absolute value of the deviation is large, such as when the deviation is greater than or equal to the preset threshold, the correction coefficient is 0.8, and the third target weight = third basic matching weight × correction coefficient. The preset threshold can be set based on empirical data or actual conditions, and is not specifically limited here.

[0146] In the above embodiments, by correcting the basic weights based on the deviation between the actual matching value and the reference value, the target weights can be dynamically adjusted according to the actual matching situation, improving the rationality of weight allocation. This dynamic correction mechanism allows the calculation of the fused scenario value to better reflect the actual situation, further improving the accuracy of business scenario identification.

[0147] In some alternative implementations, see [link to relevant documentation]. Figure 7 , Figure 7 A flowchart illustrating another method for determining business scenarios is provided, which includes the following steps:

[0148] Step 1: Obtain the user's voice information.

[0149] Step 2: Speech preprocessing.

[0150] It receives users' voice information, processes dialect-type speech through a dialect mapping library, and then converts the speech into standardized text, i.e., text information, through ASR technology, such as transcribing dialect expressions into standardized text content.

[0151] Step 3: Process the text information.

[0152] (1) It can extract keywords from text information; and for each candidate business scenario, it matches the extracted keywords with the reference words corresponding to the candidate business scenario to obtain the matching result; and based on the matching result, it obtains the keyword matching value S_keyword corresponding to the candidate business scenario.

[0153] (2) The first text vector corresponding to the text information can be obtained; for each candidate business scenario, the similarity between the first text vector and the scenario feature vector corresponding to the candidate business scenario is determined; based on the similarity and the keyword matching value corresponding to the candidate business scenario, the semantic matching value S_vector corresponding to the candidate business scenario is determined.

[0154] (3) It can obtain the user's business information; it can fuse the second text vector corresponding to the text information and the business vector corresponding to the business information to obtain the fused vector; it can input the fused vector into the scene classification model to obtain the second scene matching value P_bert corresponding to each candidate business scene output by the scene classification model.

[0155] Step 4: Merge matching values.

[0156] For each candidate business scenario, the keyword matching value S_keyword, semantic matching value S_vector, and second scenario matching value P_bert corresponding to the candidate business scenario are fused to obtain the fused scenario value.

[0157] Step 5: Output the target business scenario.

[0158] The candidate business scenario corresponding to the highest fusion scenario value is determined as the target business scenario for the user.

[0159] Compared with the prior art, the embodiments of this application have the following significant advantages and positive effects:

[0160] 1. Improved recognition accuracy and solved the problems of missed detection and false judgment.

[0161] By employing a three-tiered collaborative mechanism of "keyword matching + vector semantic matching + BERT decision-making," the limitations of single keyword matching are overcome. The vector matching layer and the BERT layer can effectively identify complex language variations such as dialects, synonyms, and antonyms, reducing the false negative rate.

[0162] By leveraging the exclusion word settings in the keyword library, the deep semantic understanding of the vector space, and the context-aware capabilities of the BERT model, it is possible to accurately distinguish the user's true intent, such as differentiating between "disassembling the device" and "fault complaint," thereby reducing the false positive rate.

[0163] The three-tiered engines can be processed in parallel, serving as backups and verifications for each other. For example, when a BERT model inference results in extremely low confidence, the results of keyword and vector matching can be referenced more extensively. This dynamic weight fusion mechanism enhances the system's robustness and avoids the risk of the entire system collapsing due to the failure of a single model.

[0164] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0165] Based on the same inventive concept, this application also provides a business scenario determination apparatus for implementing the business scenario determination method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more embodiments of the business scenario determination apparatus provided below can be found in the limitations of the business scenario determination method described above, and will not be repeated here.

[0166] In one exemplary embodiment, such as Figure 8 As shown, a business scenario determination device is provided, including:

[0167] The acquisition module 10 is used to acquire the user's voice information and convert the voice information into text information;

[0168] The matching module 20 is used to match the text information with the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value corresponding to each candidate business scenario; wherein, the first scenario matching value includes keyword matching value and / or semantic matching value;

[0169] The determination module 30 is used to determine the second scenario matching value corresponding to each candidate business scenario based on the scenario classification model, according to the text information and the user's business information;

[0170] Selection module 40 is used to select the target business scenario corresponding to the user from each candidate business scenario based on the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario.

[0171] In one embodiment, the matching module 20 is specifically used for:

[0172] Extract keywords from the text information; for each candidate business scenario, match the extracted keywords with the reference words corresponding to the candidate business scenario to obtain a matching result; based on the matching result, obtain the keyword matching value corresponding to the candidate business scenario.

[0173] In one embodiment, the matching module 20 is specifically used for:

[0174] Obtain the first text vector corresponding to the text information; for each candidate business scenario, determine the similarity between the first text vector and the scenario feature vector corresponding to the candidate business scenario; based on the similarity and the keyword matching value corresponding to the candidate business scenario, determine the semantic matching value corresponding to the candidate business scenario.

[0175] In one embodiment, the matching module 20 is specifically used for:

[0176] Obtain the user's business information; fuse the second text vector corresponding to the text information and the business vector corresponding to the business information to obtain a fused vector; input the fused vector into a scene classification model to obtain the second scene matching value corresponding to each candidate business scene output by the scene classification model.

[0177] In one embodiment, the selection module 40 is specifically used for:

[0178] For each candidate business scenario, the first scenario matching value and the second scenario matching value corresponding to the candidate business scenario are fused to obtain a fused scenario value; the candidate business scenario corresponding to the highest fused scenario value is determined as the target business scenario corresponding to the user.

[0179] In one embodiment, the selection module 40 is specifically used for:

[0180] Obtain the first target weight corresponding to the keyword matching value, the second target weight corresponding to the semantic matching value, and the third target weight corresponding to the second scenario matching value; apply the first target weight to the keyword matching value, apply the second target weight to the semantic matching value, and apply the third target weight to the second scenario matching value; determine the fusion scenario value corresponding to the candidate business scenario based on the weighting result.

[0181] In one embodiment, the selection module 40 is specifically used for:

[0182] Obtain the first basic matching weight corresponding to the keyword matching value, the second basic matching weight corresponding to the semantic matching value, and the third basic matching weight corresponding to the second scenario matching value; based on the deviation between the keyword matching value and the keyword reference value, correct the first basic matching weight to obtain the first target weight; based on the deviation between the semantic matching value and the semantic matching reference value, correct the second basic matching weight to obtain the second target weight; based on the deviation between the model prediction value and the model prediction reference value, correct the third basic matching weight to obtain the third target weight.

[0183] The modules in the aforementioned business scenario determination device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0184] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores user voice data and business scenario data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a business scenario determination method.

[0185] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0186] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the business scenario determination method described in any of the above embodiments.

[0187] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the business scenario determination method described in any of the above embodiments.

[0188] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the business scenario determination method described in any of the above embodiments.

[0189] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0190] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0191] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0192] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for determining a business scenario, characterized in that, The method includes: Acquire the user's voice information and convert the voice information into text information; The text information is matched with the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value for each candidate business scenario; wherein, the first scenario matching value includes keyword matching value and / or semantic matching value; Based on the scenario classification model, a second scenario matching value is determined for each candidate business scenario according to the text information and the user's business information. Based on the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario, the target business scenario corresponding to the user is selected from each candidate business scenario.

2. The method according to claim 1, characterized in that, The text information is matched with the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value for each candidate business scenario, including: Extract keywords from the text information; For each candidate business scenario, the extracted keywords are matched with the reference words corresponding to the candidate business scenario to obtain the matching results; Based on the matching results, the keyword matching values ​​corresponding to the candidate business scenarios are obtained.

3. The method according to claim 1 or 2, characterized in that, The text information is matched with the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value for each candidate business scenario, including: Obtain the first text vector corresponding to the text information; For each candidate business scenario, determine the similarity between the first text vector and the scenario feature vector corresponding to the candidate business scenario; Based on the similarity and the keyword matching value corresponding to the candidate business scenario, the semantic matching value corresponding to the candidate business scenario is determined.

4. The method according to claim 1, characterized in that, Based on the scenario classification model, and according to the text information and the user's business information, a second scenario matching value is determined for each candidate business scenario, including: Obtain the user's business information; The second text vector corresponding to the text information and the business vector corresponding to the business information are fused to obtain a fused vector; The fusion vector is input into the scene classification model to obtain the second scene matching value corresponding to each candidate business scene output by the scene classification model.

5. The method according to claim 1, characterized in that, The step of selecting the target business scenario corresponding to the user from each candidate business scenario based on the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario includes: For each candidate business scenario, the first scenario matching value and the second scenario matching value corresponding to the candidate business scenario are fused to obtain the fused scenario value; The candidate business scenario corresponding to the highest fusion scenario value is determined as the target business scenario for the user.

6. The method according to claim 5, characterized in that, The step of fusing the first scenario matching value and the second scenario matching value corresponding to the candidate business scenario to obtain the fused scenario value includes: Obtain the first target weight corresponding to the keyword matching value, the second target weight corresponding to the semantic matching value, and the third target weight corresponding to the second scene matching value; The keyword matching value is weighted using the first target weight, the semantic matching value is weighted using the second target weight, and the second scene matching value is weighted using the third target weight. Based on the weighted processing results, the fusion scenario value corresponding to the candidate business scenario is determined.

7. The method according to claim 6, characterized in that, The step of obtaining the first target weight corresponding to the keyword matching value, the second target weight corresponding to the semantic matching value, and the third target weight corresponding to the second scene matching value includes: Obtain the first basic matching weight corresponding to the keyword matching value, the second basic matching weight corresponding to the semantic matching value, and the third basic matching weight corresponding to the second scenario matching value; Based on the deviation between the keyword matching value and the keyword reference value, the first basic matching weight is corrected to obtain the first target weight; Based on the deviation between the semantic matching value and the semantic matching reference value, the second basic matching weight is corrected to obtain the second target weight; Based on the deviation between the model's predicted value and the model's predicted reference value, the third basic matching weight is corrected to obtain the third target weight.

8. A business scenario determination device, characterized in that, The device includes: The acquisition module is used to acquire the user's voice information and convert the voice information into text information; The matching module is used to match the text information with the scenario information corresponding to each candidate business scenario to obtain a first scenario matching value corresponding to each candidate business scenario; wherein, the first scenario matching value includes keyword matching value and / or semantic matching value; The determination module is used to determine the second scenario matching value corresponding to each candidate business scenario based on the scenario classification model, according to the text information and the user's business information; The selection module is used to select the target business scenario corresponding to the user from each candidate business scenario based on the first scenario matching value and the second scenario matching value corresponding to each candidate business scenario.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images